On August 20, 2025, Luke W. Russell, an Indianapolis filmmaker, finds themselves turning to ChatGPT for emotional support. (Luke uses they/them pronouns.) “I have cried while working through things with it,” Russell says, describing how they use the chatbot to process intrusive thoughts or navigate traumatic memories.
“I have had countless experiences where what ChatGPT said to me felt so true, so powerful, that I felt profoundly seen.” Russell’s experience reflects a broader, growing reality where many individuals are increasingly turning to chatbots for mental health support. This ranges from managing anxiety and processing grief to navigating work conflicts and resolving marital disputes. According to a 2025 Harris Poll and American Psychological Association (APA) survey, over half of adults aged 18 to 54 and a quarter of those 55 and older expressed willingness to discuss their mental health with an AI chatbot.
The Critical Gap Between Design and Real-World Use
A significant problem underlies this trend. AI chatbots like OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini were not designed for mental health therapy. Even AI products marketed as emotional wellness tools, such as Replika, Wysa, Youper, and MindDoc, are not built on validated psychological methods. Dr. C. Vaile Wright, a psychologist and senior director at the APA’s Office of Health Care Innovation, states she does not believe any commercially available AI-assisted therapy is truly approved yet. She points out that many existing chatbots lack foundational research, psychological science, and input from subject matter experts. This design gap raises serious concerns about their application in sensitive mental health contexts.
Potential Risks and the Push for Regulatory Oversight
Critics warn that potential AI biases, a lack of genuine empathy, and limited human oversight could actively jeopardize user mental health. This is particularly concerning for vulnerable groups like children, adolescents, individuals with pre-existing mental health conditions, and those experiencing suicidal thoughts. Growing concerns have led to terms like “ChatGPT psychosis” or “AI psychosis,” referring to the potential harmful psychological effects of AI interactions. This issue has captured legislative attention.
Illinois recently enacted restrictions on AI in mental health care, prohibiting its use for treatment and banning mental health professionals from using AI to communicate with clients or make treatment decisions. Similar measures have been passed in Nevada and Utah. However, these regulations do not prevent individuals from seeking support from chatbots, especially amid a shortage of clinicians, rising therapy costs, and inadequate mental health insurance coverage.
The Driving Forces Behind the Turn to AI for Support
The appeal of chatbots is largely driven by systemic gaps in traditional mental health care. Data reveals a critical shortage of mental health workers, especially in remote and rural areas, says Dr. Elizabeth Stade, a research fellow at Stanford University’s Computational Psychology and Well-Being Lab. Dr. Jessica Schleider, a child and adolescent psychologist at Northwestern University, adds that among U.S. adults with significant mental health needs, only about half access any form of treatment; for youth, the figure is closer to 75%.
“This provider shortage is clearly driving many to turn to their devices, and increasingly to generative AI, to fill this void.” Unlike therapists, chatbots are available 24/7, providing support during off-hours, which is often when people need help most. Dr. Wright suggests that properly designed AI tools could potentially supplement human therapy.
Early Research Findings and User-Reported Benefits
Initial clinical trial results for AI-generated therapy chatbots show promise. Dr. Michael V. Heinz, lead author of a study and a professor at Dartmouth’s Geisel School of Medicine, reported “significant reductions in symptoms of depression, anxiety, and eating disorders” over four to eight weeks using a chatbot called Therabot, which combines evidence-based interventions with advanced generative AI. Participants reported high engagement levels and described the experience as similar to talking to a human therapist, though Dr. Heinz cautions that results are early and more research is needed. For users like Russell, access and affordability are key attractions.
“I didn’t intend to use ChatGPT as a therapist,” Russell explains, having stopped traditional therapy due to income decline. The non-judgmental nature of chatbots is also a significant draw. Dr. Jonathan Gratch from the University of Southern California notes that people are reluctant to be judged, making them more willing to disclose symptoms to a virtual agent, a finding supported by his research with veterans.
Ethical Concerns and the Future of AI in Mental Health
Despite potential benefits, serious ethical concerns persist. Dr. Henry Willis, a psychologist at the University of Maryland, points out that most users do not understand how AI works, often assuming it is always objective. AI training data is frequently unrepresentative of minority groups, leading to biases and technologically-mediated racism. Chatbots can also make “ecological fallacies,” treating individual comments as data points and making assumptions based on group data that may not reflect personal reality, explains Dr. H. Andrew Schwartz of Stony Brook University.
There are also safety and accountability issues, with some chatbots linked to cases involving violence and self-harm. Dr. Wright expresses concern about chatbots designed for profit, which may unconditionally validate users—even if their responses are harmful—to keep them engaged. Furthermore, these conversations are not protected by HIPAA regulations, leaving personal information vulnerable.
The psychological impact of forming emotional bonds with AI is another area of concern, potentially replacing real human relationships and reducing motivation for social interaction, which could increase loneliness long-term. The future, according to Dr. Wright, lies in well-developed, scientifically-driven chatbots with proper safety safeguards for crisis situations, but she concludes, “We are not quite there yet.”
Related topics
